Aerial Image Classification Using Deep Learning and Machine Learning Models

Aerial Image Classification plays a pivotal role in environmental monitoring, urban planning, and resource management. With the rise of Machine Learning and Deep Learning, automated scene interpretation has achieved remarkable precision. This study presents an experimental comparison between classical machine learning algorithms and deep learning models for classifying aerial imagery using the UC Merced Land Use Dataset, which contains 21 land-use categories of high-resolution images. After preprocessing through resizing, normalization, and feature extraction using Histogram of Oriented Gradients (HOG) and Convolutional Neural Network (CNN) embeddings, four models Random Forest, Support Vector Machine, a custom CNN, and a fine-tuned Transfer Learning (VGG16) were evaluated. The results demonstrate that deep learning approaches significantly outperform traditional methods, with the fine-tuned CNN achieving 91.2% accuracy, confirming the superior capability of CNN-based feature extraction for spatial scene analysis and emphasizing the feasibility of deep learning research with limited computational resources.